Fix it now
Error 8531 is ERROR_DS_CANT_START, and Microsoft says exactly that much: Directory Service cannot start. What went wrong is in the codes logged alongside it. Read those first, because the answers they point at range from a full volume to a database that has to be restored, and the wrong first move makes the second impossible.
Get-WinEvent -LogName "Directory Service" -MaxEvents 60 | Format-List TimeCreated,Id,LevelDisplayName,Message
Get-Volume | Format-Table DriveLetter,FileSystemLabel,SizeRemaining,Size
- Read the Directory Service entries around the failure. 8541 is a Security Accounts Manager initialisation failure, which is what reboots a controller into Safe Mode; 8451 and 8443 are replication hitting a database error or a database inconsistency; 8573 is the database out of version store, which is a transaction problem rather than a disk problem.
- Check free space on the volumes holding the database and the log files – which are not always C, and are recorded under the NTDS Parameters registry key. Free space by moving other data off the volume.
- Take a copy of the whole NTDS folder somewhere else before you do anything more. Everything below is easier to undo if you can put the current state back.
- Decide whether you have another healthy domain controller. If you do, stop troubleshooting this one: force its removal with
Uninstall-ADDSDomainController -ForceRemovalrun on the failed server, clean its metadata out from a healthy controller, and promote it again.
Never delete the directory’s transaction log files to free disk space. They are the route back to a consistent database after an unclean shutdown, and removing them converts a recoverable server into a restore from backup.
If the service starts and stays started, you are done. If it does not, the next section explains what each of the companion codes is telling you and which of them changes the plan.
Why it happens
8531 on its own is not a diagnosis. Microsoft publishes it as ERROR_DS_CANT_START, 0x2153, “Directory Service cannot start”, and nothing more. Every article that tells you what 8531 means is telling you what it usually meant on the machine the author was looking at. The information is in the entries logged with it, and those do have exact published meanings.
8541 is ERROR_SAM_INIT_FAILURE: Security Accounts Manager initialisation failed with a given error status, and the system will shut down and reboot into Safe Mode. On a domain controller that is the observable event – the server comes up in Directory Services Restore Mode and nobody can sign in with a domain account. It is a symptom of the directory failing to start, not a separate fault.
8451 is ERROR_DS_DRA_DB_ERROR and 8443 is ERROR_DS_DRA_INCONSISTENT_DIT: the replication operation encountered a database error, or a database inconsistency. Both are replication reporting that the storage underneath it is unhappy, which is a different statement from the network or the security layer being unhappy – and it means repairing replication is not the work.
8573 is the one most often misread. ERROR_DS_OUT_OF_VERSION_STORE means the database is out of version store, which is a limit on how much uncommitted change a single transaction may hold. It is produced by one very long operation – a huge group membership change, a mass delete, a badly written LDAP client – and not by a full disk. Free space will not fix it and neither will a reboot on its own; the operation has to be broken into smaller pieces.
The most important judgement is not about any of those codes. It is whether this domain controller is the only copy of something. If another healthy controller holds the same partitions, the data is safe, this machine is a broken replica, and rebuilding it is faster and lower risk than any database work. Database repair belongs to the case where this really is the last copy.
The volume holding the database or the logs is full
You have this one if Free space at or near zero on the volume named in the NTDS Parameters key, and the failure appeared without anyone changing anything.
- Free space by moving something that is not part of the directory: old backups, dump files, installer caches, or a page file sharing the volume.
- Start the AD DS service, confirm the directory comes online, and set monitoring on that volume so the next occurrence is a warning rather than an outage.
An offline defragmentation reclaims space inside the database file but needs free space of its own to write the compacted copy. It is something you do after you have room, not a way of making room.
Another healthy domain controller holds the same partitions
You have this one if The domain has at least one other controller that is replicating normally with the rest of the estate.
- Copy the NTDS folder off the failed server first, so you can return to the current state.
- Run
Uninstall-ADDSDomainController -ForceRemovalon the failed server. Microsoft describes that parameter as forcing the uninstall where you do not have connectivity to other controllers in the topology. - From a healthy controller, clean the failed server’s metadata out of the directory and tidy its records in Sites and Services and DNS.
- Promote the rebuilt server again and confirm replication converges.
This is the answer far more often than the database repair articles suggest. A domain controller is a replica; a broken replica is a rebuild, not a recovery project.
The database is out of version store
You have this one if 8573 in the log, and the failure lines up with a bulk operation – a large deletion, a group membership change, a synchronisation tool run.
- Identify the operation. It is usually the only unusual thing that happened in the window.
- Break it into smaller transactions and run it in batches rather than as one pass. Do not treat it as a disk-space or corruption problem; the volume can be half empty and the database perfectly healthy.
This is the only domain controller and there is a backup
You have this one if No other controller holds the domain, and you have a system state backup from before the failure.
- Copy the NTDS folder off the machine as it stands, before restoring anything over the top of it.
- Restore the system state from the most recent backup you trust, then confirm the directory service starts and that recent changes you expected are present.
- Once it is running, build a second domain controller so the next failure is an incident rather than an outage.
Antivirus or backup software is holding the database files
You have this one if The failure is intermittent, clusters around a backup window, and the database is otherwise healthy.
- Exclude the database path, the log path and the working directory from on-access scanning, and confirm the backup product uses a directory-aware method rather than a file copy of an open database.
- Restart and watch across several backup cycles before calling it fixed, because one clean night proves nothing here.
Full reference
What each code alongside 8531 actually says
| Code | Name | Published meaning |
|---|---|---|
| 8531 | ERROR_DS_CANT_START | Directory Service cannot start |
| 8541 | ERROR_SAM_INIT_FAILURE | Security Accounts Manager initialisation failed with the given error status; shut down and reboot into Safe Mode |
| 8451 | ERROR_DS_DRA_DB_ERROR | The replication operation encountered a database error |
| 8443 | ERROR_DS_DRA_INCONSISTENT_DIT | The replication operation encountered a database inconsistency |
| 8573 | ERROR_DS_OUT_OF_VERSION_STORE | The database is out of version store |
Where the database and its logs actually live
Read the paths rather than assuming the defaults. They are recorded under the NTDS service’s Parameters key, at HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\NTDS\Parameters, and a controller built by somebody else may have them on different volumes from the ones you are watching. A full volume you are not looking at is a common reason this arrives unannounced.
Never delete the transaction log files to free disk space. They are the only route back to a consistent database after an unclean shutdown, and removing them turns a routine recovery into a restore from backup – or into permanent loss where the backup is missing too. Free space by moving something else off the volume.
Things that make the situation worse
- Rolling a domain controller back from a virtual machine snapshot or a whole-image restore. That reuses replication sequence numbers the rest of the domain has already seen, and the inconsistency is harder to resolve than the failure you started with.
- Running a repair tool against the live database before taking a copy. A repair that discards what it cannot read cannot tell you what it discarded.
- Repromoting over the top. The role cannot be removed cleanly from a machine whose database will not open, so this produces a second problem rather than solving the first.
- Deleting log files, checkpoint files or anything else in the NTDS folder to make room.
- Bringing the machine back onto the network after a forced removal without cleaning its metadata, which reintroduces objects the rest of the domain has already deleted.
The order that keeps your options open
- Copy the NTDS folder somewhere safe. Everything after this is reversible only if you did this.
- Read the Directory Service log and identify which companion codes are present.
- Check free space on the volumes named in the NTDS Parameters key, and free space by moving unrelated data.
- Establish whether another healthy domain controller holds the same partitions. If it does, rebuild this one rather than repairing it.
- If it does not, restore the system state from the most recent backup you trust.
- Treat any in-place database repair as a way of extracting data from a machine you will not keep, rather than as a way of returning a server to service.
After it is running again
- Confirm the AD DS service starts on a normal boot, not only when started by hand.
- Run the built-in domain controller diagnostics and confirm the core tests pass on this server.
- Check inbound and outbound replication with its partners and let a full cycle complete before you close the incident.
- Look at what filled the volume or ran the long transaction, because unaddressed it will happen again on the same schedule.
- Count your domain controllers. A single-controller domain turns every one of these into a full outage.
When a licence is the actual fix
The work itself costs nothing: the diagnostics, the forced removal, the metadata cleanup and Windows Server Backup all ship with the product, and the honest order of operations is free disk space, then rebuild from a healthy replica, then restore. The purchase case sits next to the problem rather than inside it. If this outage stopped authentication for the whole organisation, it did so because there was only one domain controller, and a second turns a database failure into an inconvenience. Arco supplies Windows Server 2025 Standard, whose licence carries the right to run two virtual machines plus one Hyper-V host, so a second virtual controller may already be covered by capacity you hold – we would rather check that than sell you something you do not need. If you already run two or more healthy controllers, there is nothing to buy here.
Every code this article covers
| Code | What it points at | Source |
|---|---|---|
8531 |
ERROR_DS_CANT_START, 0x2153: Directory Service cannot start. It carries no detail of its own; the entries logged with it do | Microsoft Learn |
8541 |
ERROR_SAM_INIT_FAILURE, 0x215D: Security Accounts Manager initialisation failed with the stated error status, and the system reboots into Safe Mode | Microsoft Learn |
8451 |
ERROR_DS_DRA_DB_ERROR, 0x2103: the replication operation encountered a database error | Microsoft Learn |
8443 |
ERROR_DS_DRA_INCONSISTENT_DIT, 0x20FB: the replication operation encountered a database inconsistency | Microsoft Learn |
8573 |
ERROR_DS_OUT_OF_VERSION_STORE, 0x217D: the database is out of version store, which is a transaction size limit rather than a disk or corruption problem | Microsoft Learn |
Event ID 1393 |
A Directory Service log entry seen when updates to the directory database are failing. Microsoft publishes no definition page for it, so read it as a pointer at the database rather than as a diagnosis | not published by the vendor |
Confirm the fix worked
- The Active Directory Domain Services service starts and stays running after a normal boot, not just when started manually.
- The Directory Service log records no further 8531, 8541, 8451, 8443 or 8573 entries over a full day.
- Inbound and outbound replication with every partner is succeeding.
- The domain controller diagnostics pass their core tests on this server.
- Free space on the database and log volumes is being monitored, so the same cause cannot arrive silently again.
Questions people ask about this
Can I just repromote the domain controller?
Not while the service will not start – the role cannot be removed cleanly from a machine whose database will not open. The route is a forced removal on the machine itself, metadata cleanup from a healthy controller, then promotion again.
How much free space does the directory need?
There is no single number worth quoting, because it depends on the size of the database and what maintenance you run. Treat the volume as the production database volume it is, monitor it, and do not let it approach full.
Is my data lost?
Usually not. If another domain controller holds the same partitions, everything is still there and this machine is only a broken copy. The dangerous case is a single domain controller with no backup, which is a design problem that this failure has merely exposed.
Should I repair the database before restoring from backup?
No. Restore first if you have a good backup. A repair discards whatever it cannot read and cannot tell you what that was, so it is what you do when there is no backup and no other replica – and even then, treat the result as a way of extracting data rather than a server you keep.
8573 keeps coming back after a reboot. Why?
Because it is not a disk or corruption problem. Out of version store is a limit on a single transaction, so it returns every time the same oversized operation runs. Break the operation into batches.
